Last Updated on March 19, 2025 by Editor
This AI Week # 11 (March 12–18, 2025) has been a whirlwind of game-changing innovations, billion-dollar investments, and major breakthroughs that you can’t afford to miss! Google, Meta, OpenAI, Nvidia, Anthropic, and other AI giants are pushing boundaries—Google is integrating Gemini into Assistant, Meta’s Llama models have surpassed 1 billion downloads, and Nvidia is collaborating on AI-native 6G networks. Meanwhile, startups like Halliday, ReactWise, and Sesame are securing millions to revolutionize blockchain security, drug discovery, and virtual assistants. Snapchat’s AI-powered video lenses, Roblox’s 3D object creation model, and Xbox’s AI gaming companion are reshaping digital experiences. And let’s not forget Elon Musk’s xAI acquiring a generative AI video startup and SoftBank investing $676M in AI infrastructure in Japan.
With open-source models outperforming GPT-4, AI-powered cybersecurity strategies, and robotics breakthroughs, this is the future happening NOW. Stay ahead of the curve or risk being left behind—because AI isn’t waiting for anyone!
Stability AI Unveils Stable Virtual Camera to Transform 2D Images into Immersive 3D Scenes
Stability AI has introduced the Stable Virtual Camera, a cutting-edge AI model designed to convert 2D images into immersive 3D videos. This multi-view diffusion model allows users to generate videos with realistic depth and perspective, offering dynamic camera paths such as Spiral, Dolly Zoom, and 360° views. The tool supports up to 32 input images to generate “novel views” and produces outputs in square (1:1), portrait (9:16), and landscape (16:9) aspect ratios, with a maximum length of 1,000 frames.
The model is currently in a research preview and available under a non-commercial license. While it excels in generating smooth 3D visuals, challenges remain with complex scenes or irregularly shaped objects that may cause flickering artifacts. Stability AI invites researchers to explore its potential through platforms like Hugging Face. This release follows a series of initiatives by Stability AI, including partnerships and new AI models aimed at revitalizing the company after prior financial setbacks.
Google Enhances Gemini with Canvas Feature and Audio Overview
Google has expanded its Gemini AI chatbot with two innovative features: Canvas and Audio Overview. Canvas provides an interactive workspace for creating and refining writing or coding projects, enabling users to edit drafts, adjust tone, or preview code in real-time. For instance, users can ask Gemini to generate HTML for an email subscription form and preview its functionality instantly.
Additionally, the Audio Overview feature generates podcast-style summaries of documents or webpages, offering a new way to consume information. Both features are accessible via the Gemini app on web and mobile platforms. These updates position Gemini as a robust productivity suite, enhancing collaboration and efficiency for both personal and professional use cases.
Adobe Launches AI Agents to Revolutionize Customer Experience
Adobe has introduced the Adobe Experience Platform Agent Orchestrator, enabling businesses to deploy AI agents for personalized customer interactions. These agents can optimize websites, refine marketing workflows, and scale digital media production. Notably, Adobe’s Brand Concierge application allows brands to configure agents tailored to their unique identity, providing immersive conversational experiences.
The platform integrates with major partners like Microsoft and Amazon Web Services, ensuring seamless execution across ecosystems. With AI-driven tools that analyze user behavior—such as tailoring content based on ad origin—Adobe aims to accelerate tasks that previously required months of development. This innovation underscores Adobe’s commitment to enhancing digital marketing efficiency.
LG Introduces Exaone Deep for Advanced Reasoning in Math, Science, and Coding
LG unveiled Exaone Deep, a 32-billion parameter AI model specializing in logical problem-solving at Nvidia’s GTC 2025 conference. The model excels in tasks requiring deep reasoning, such as solving graduate-level math problems and coding challenges. It scored an impressive 94.5 on Korea’s CSAT math exam and 90 on the AIME 2024, showcasing its advanced capabilities.
Exaone Deep employs structured logic on top of pattern recognition to reduce random outputs while clarifying reasoning steps. Open-sourced for further development, this model positions LG as a competitive player in inference AI technology globally.
Graphite Secures $52M Series B to Revolutionize AI-Powered Code Review
Graphite, an AI-powered code review platform, has raised $52 million in a Series B funding round, led by Accel, with participation from Anthropic’s Anthology Fund, Menlo Ventures, Shopify Ventures, Figma Ventures, and Andreessen Horowitz. This brings Graphite’s total funding to $81 million since its founding in 2020 by Tomas Reimers, Greg Foster, and Merrill Lutsky. Initially focused on mobile development tools, Graphite shifted to AI-driven code review and now serves over 500 companies, including Shopify and Snowflake.
The platform uses Anthropic and OpenAI models to detect errors, propose fixes, and summarize code. In 2024 alone, Graphite reported a 20x revenue growth, driven by its adoption across engineering teams. To expand its offerings, the company is launching Diamond, a standalone tool for automatic bug detection. The funds will be used for product development and team expansion in New York City. Graphite’s unique ability to customize code patterns and protect sensitive data sets it apart in the competitive AI coding assistant market dominated by GitHub Copilot and others.
Arcade Raises $12M to Build Better AI Agent Infrastructure
Arcade, an AI agent infrastructure startup co-founded by Alex Salazar and Sam Partee in 2024, has raised $12 million from Laude Ventures, a fund led by Perplexity co-founder Andy Konwinski. Arcade aims to address the inefficiencies of current AI agents, which often fail to integrate with private data or perform reliably. Instead of building agents directly, Arcade provides a tool-calling platform that allows agents to securely access apps and data with user-level permissions.
The platform integrates with OAuth for seamless authentication across thousands of SaaS services while ensuring security through token management. This approach enables agents to perform tasks like site reliability monitoring more effectively. Arcade offers usage-based pricing or subscriptions and is designed for developers seeking robust AI agent infrastructure. The funding will help Arcade refine its platform and expand its reach in the growing AI agent market.
Mistral Small 3.1: A Lightweight Open-Source Model Outperforming GPT-4o Mini
French startup Mistral AI has released Mistral Small 3.1, a 24-billion-parameter open-source model that surpasses Google’s Gemma 3 and OpenAI’s GPT-4o Mini in benchmarks like MMLU (general knowledge), MATH (reasoning), and GPQA (question answering). Despite its smaller size, the model boasts a 128K token context window, making it ideal for long-form reasoning tasks such as document analysis.
Unlike many proprietary models requiring expensive cloud infrastructure, Mistral Small 3.1 can run locally on hardware like an RTX 4090 or a MacBook with 32GB RAM when quantized. Released under an Apache 2.0 license, it is freely available for developers to fine-tune for specific applications in fields like healthcare and legal services. This positions Mistral as a strong competitor against closed-source giants like OpenAI while promoting accessibility in the AI ecosystem.
Meta’s Llama Models Surpass 1 Billion Downloads Amid Rapid Expansion
Meta CEO Mark Zuckerberg announced that downloads of the Llama series of AI models have exceeded 1 billion, marking a significant increase from 650 million in December 2024—a growth of approximately 53% in just three months. Meta plans to invest up to $80 billion in AI projects this year, including the release of new Llama models with multimodal capabilities and inference functionalities similar to OpenAI’s offerings.
The upcoming LlamaCon developer conference on April 29 is expected to unveil further details about these advancements. Meta’s aggressive push into the AI space underscores its ambition to compete with industry leaders like OpenAI while fostering innovation through open-source initiatives.
NVIDIA Partners with Telecom Leaders to Build AI-Powered 6G Wireless Networks
NVIDIA has announced a groundbreaking initiative to develop AI-native 6G wireless networks, partnering with global telecom leaders such as T-Mobile and Booz Allen Hamilton. This announcement was made at the GTC 2025 conference, where NVIDIA emphasized the integration of AI into the core architecture of next-generation wireless systems. The collaboration aims to create software-defined radio access networks (RAN) powered by NVIDIA’s accelerated computing infrastructure, enabling unprecedented spectral efficiency—a critical metric for data transmission rates over limited bandwidth.
These AI-driven networks are expected to connect billions of devices, ranging from smartphones and sensors to autonomous vehicles, while optimizing performance and resource management. T-Mobile’s CEO, Mike Sievert, highlighted the potential for AI-enhanced scalability and efficiency, paving the way for revolutionary user experiences. NVIDIA’s Ronnie Vasishta noted that 95% of telecom companies are already engaging with NVIDIA’s solutions, underscoring the industry’s shift toward AI-integrated networks. With these advancements, 6G networks could support hundreds of billions of devices globally, setting a new standard for connectivity and innovation.
Anthropic Prepares Two-Way Voice Mode for Claude AI
Anthropic is set to roll out a two-way voice interaction feature for its AI chatbot, Claude, enabling users to engage in natural, hands-free conversations. This enhancement will allow Claude to both listen and respond vocally, making interactions more fluid and human-like. In addition to voice capabilities, Anthropic is introducing a memory feature that enables Claude to recall past interactions, tailoring responses based on user preferences and previous conversations. For example, if a user mentions a favorite book or hobby, Claude can reference it in future dialogues, creating a more personalized experience.
These updates are part of Anthropic’s broader strategy to compete with AI giants like OpenAI and Google by focusing on more intuitive and human-centric technologies. However, concerns about data privacy and memory accuracy remain critical as these features roll out in the coming months. By addressing these challenges, Anthropic aims to position Claude as a leading conversational AI assistant in an increasingly competitive market.
Google Adds New Healthcare Features to Search and Android
Google has introduced a suite of healthcare-focused features across its platforms, including enhanced responses in Search for health-related queries and APIs for medical records on Android devices. Using advanced AI ranking systems, Google has expanded its “knowledge panel” answers to cover thousands of health topics in languages like Spanish, Portuguese, and Japanese. Additionally, a new U.S.-based feature called “What People Suggest” aggregates user-generated insights on health conditions from forums and online communities.
In parallel, Google launched APIs under its Health Connect platform that allow apps to read and write medical data such as immunizations and lab results using standardized formats like FHIR (Fast Healthcare Interoperability Resources). Furthermore, Google unveiled TxGemma, an open AI model designed for drug discovery tasks, following its earlier Gemini models tailored for healthcare applications. These innovations underscore Google’s commitment to leveraging AI for improving global healthcare accessibility and personalization.
Halliday Secures $20M Series A Funding for Safe Blockchain-Based AI Agents
Halliday, an innovator in blockchain-based AI agents, has raised $20 million in Series A funding, led by Andreessen Horowitz’s crypto arm (a16z crypto). The company’s core innovation is the Agentic Workflow Protocol (AWP), which establishes “immutable guardrails” for AI agents operating within decentralized environments. This ensures that autonomous systems execute tasks securely within predefined limits—critical for enterprise applications involving financial systems or digital assets.
Halliday has already deployed its workflow engine with partners like DeFi Kingdoms and ApeChain, automating complex processes such as treasury management and recurring payments across multiple blockchain networks. By reducing development costs by up to 10,000x, Halliday aims to make AI-blockchain integration accessible even to non-specialized teams. With this funding round bringing its total investment to over $26 million, Halliday plans to expand its protocol development and team while addressing key challenges in safety and compliance for blockchain-based AI solutions.
Google Introduces TxGemma AI Models to Revolutionize Drug Discovery
At its “The Check Up” event in New York, Google unveiled TxGemma, a suite of AI models for drug discovery designed to streamline early-stage therapeutic development. These open AI models, part of Google’s Health AI Developer Foundations program, can analyze the structures of chemicals, molecules, and proteins while answering critical research questions about potential therapies. For example, researchers can use TxGemma to predict the safety and efficacy of new drugs, significantly reducing the time and cost associated with traditional R&D processes.
Karen DeSalvo, Google’s Chief Health Officer, emphasized that TxGemma is aimed at addressing the inefficiencies in drug development, which often takes years and billions of dollars. While specific licensing details for commercial use remain unclear, the models are expected to be released later this month. This initiative builds on Google’s broader healthcare ambitions, including its Gemini-based AI co-scientist tools and Health Connect APIs for medical data integration. With over 460 startups and $60 billion invested globally in AI-driven drug discovery, Google aims to position itself as a leader in this transformative field.
Elon Musk’s xAI Acquires HotShot to Advance Generative AI Video Technology
xAI, Elon Musk’s artificial intelligence company, has acquired HotShot, a startup specializing in generative AI video tools. HotShot’s technology focuses on text-to-video generation, with three foundational models—HotShot XL, Act One, and HotShot—already developed. The acquisition aligns with Musk’s vision to integrate video-generation capabilities into xAI’s Grok chatbot platform, potentially launching a “Grok Video” model in the coming months.
HotShot CEO Aakash Sastry announced that existing users can download their content until March 30 as the platform phases out its standalone services. The startup had previously attracted funding from notable investors like Alexis Ohanian and SV Angel but did not disclose funding specifics. By leveraging xAI’s Colossus supercomputing cluster, this acquisition positions xAI to compete with platforms like OpenAI’s Sora and Google’s Veo 2 in education, entertainment, and productivity markets. This move underscores Musk’s intent to expand xAI’s generative AI capabilities beyond text into multimedia applications.
OpenAI Tests ChatGPT Connectors for Seamless Integration with Slack and Google Drive
OpenAI is beta-testing ChatGPT Connectors, a feature that allows business customers to integrate applications like Slack and Google Drive directly with ChatGPT. This functionality enables users to query files, presentations, spreadsheets, and Slack conversations within their workspace using natural language commands. Built on OpenAI’s GPT-4o model, ChatGPT Connectors aim to streamline workplace productivity by offering tailored responses based on internal company data while respecting access permissions for private files and channels.
The beta rollout includes plans to expand compatibility to platforms like Microsoft SharePoint and Box in the future. OpenAI has emphasized robust encryption measures to ensure data security during synchronization. While some organizations remain cautious about sensitive data exposure, others see this as a game-changing feature for enterprise search and collaboration tools. ChatGPT Connectors could challenge existing solutions like Glean by providing seamless access to both internal and external knowledge bases through a single interface.
YC-Backed ReactWise Uses AI to Accelerate Drug Manufacturing by 30x
ReactWise, a Cambridge-based startup backed by Y Combinator (YC), is applying AI to revolutionize chemical manufacturing processes in drug development. By leveraging an “AI copilot for chemical process optimization,” ReactWise reduces trial-and-error experimentation cycles by up to 30 times, significantly shortening drug production timelines. The company has conducted thousands of chemical reactions in its labs to train its proprietary models, which provide real-time recommendations for efficient manufacturing workflows.
With $3.4 million in pre-seed funding from YC, Innovate UK, and angel investors, ReactWise is currently running 12 pilot trials with pharmaceutical companies, including major players in Big Pharma. Its AI-driven approach could cut process development times by up to 60%, making drug manufacturing faster, cheaper, and more sustainable. Positioned as a leader in AI-driven pharmaceutical innovation, ReactWise plans full-scale deployments later this year while continuing its focus on high-quality datasets for small-molecule drugs and polymer delivery systems.
Roblox Launches Cube 3D: Open-Source AI Model for 3D Object Creation
Roblox has officially released Cube 3D, an open-source AI-powered model designed to generate 3D objects using text prompts. Currently in beta, the tool allows creators to input commands like “/generate a motorcycle” or “/generate orange safety cone,” and within seconds, fully compatible 3D models are produced. These objects can be further refined using Roblox Studio, making it ideal for game developers and virtual world creators. The model is trained on native 3D data from the Roblox platform, enabling it to tokenize shapes similarly to how large language models (LLMs) process words.
Cube 3D’s debut includes the Mesh Generation API, which simplifies asset creation and integrates seamlessly with game engines. Future updates will expand its capabilities to include text-to-speech, speech-to-text, and image-based 3D generation. By making Cube open-source, Roblox extends its utility beyond its platform, empowering developers globally. With this innovation, Roblox aims to lead in AI-driven immersive content creation while fostering creativity among its 70 million monthly active users.
Google Adds Chirp 3 Voice Model to Vertex AI Platform
Google has integrated Chirp 3, its latest speech-to-text and HD text-to-speech model, into the Vertex AI platform, enhancing voice-based AI development for applications like voice assistants, audiobooks, customer support agents, and video voiceovers. Chirp 3 supports 31 languages and offers eight new voices, making it a versatile tool for global developers. The announcement was made at Google’s DeepMind offices in London as part of its push to dominate the rapidly growing AI voice technology sector.
To address concerns about misuse, Google has implemented strict usage restrictions for Chirp 3 while emphasizing its potential for ethical applications. This addition complements other cutting-edge tools in Google’s ecosystem, such as the Gemini language model and Imagen image-generation platform. With competitors like ElevenLabs and Sesame making strides in hyper-realistic voice generation, Google’s move underscores its commitment to staying ahead in AI voice innovation while ensuring safety and accessibility for developers worldwide.
Baidu Unveils Ernie 4.5 and Ernie X1: Multimodal AI Models with Advanced Capabilities
Baidu has launched two new AI models: Ernie 4.5, an upgraded foundational model with multimodal capabilities, and Ernie X1, a reasoning-focused model designed for cost-effective performance. Ernie 4.5 excels in understanding complex inputs like memes and satire, while Ernie X1 delivers results comparable to DeepSeek R1 but at half the cost. Both models can process text, images, video, and audio inputs, positioning Baidu as a leader in multimodal AI innovation.
These advancements come as Baidu prepares for the release of Ernie 5 later this year, which promises further improvements in multimodal reasoning. Despite challenges in gaining widespread adoption compared to rivals like OpenAI’s GPT-4o, Baidu’s focus on affordability and versatility could make these models attractive for businesses seeking high-performance AI solutions without breaking the bank.
Google Replaces Assistant with Gemini: A New Era of AI-Powered Personal Assistants
Google has officially announced that it will replace Google Assistant with its advanced Gemini AI assistant across devices by late 2025. This transition marks a significant shift toward generative AI-powered personal assistants capable of interacting with apps, managing tasks, and providing personalized recommendations. Gemini is already available in over 40 languages across 200 countries, with millions of users having voluntarily adopted it since its initial launch in February 2024.
The replacement extends beyond smartphones to tablets, smartwatches, cars, and home devices like Nest Hubs and TVs. Devices running Android 9 or earlier or with less than 2GB of RAM will retain Google Assistant due to compatibility issues. Gemini promises enhanced productivity and creativity features while integrating seamlessly with Google’s broader ecosystem of AI tools such as Vertex AI and Imagen. This move underscores Google’s ambition to redefine digital assistance through cutting-edge generative AI technology.
SoftBank Acquires Former Sharp Plant for $677 Million to Boost OpenAI Collaboration in Japan
In a strategic move to enhance its artificial intelligence capabilities, SoftBank has acquired a former Sharp LCD panel plant in Osaka, Japan, for approximately 100 billion yen ($677.05 million). This facility is set to be transformed into a state-of-the-art AI data center in collaboration with OpenAI. Scheduled to commence operations in 2026, the center will feature a substantial power capacity of 150 megawatts, positioning it among Japan’s largest data centers. The initiative aims to commercialize OpenAI’s agent models by training them on client data, offering customized AI solutions to businesses. The total investment for this ambitious project could escalate to 1 trillion yen ($6.77 billion), underscoring SoftBank’s commitment to advancing AI infrastructure in the region. citeturn0news42
Sesame Unveils Open-Source AI Model Powering Virtual Assistant ‘Maya’
Sesame, the innovative startup behind the viral virtual assistant ‘Maya’, has released its foundational AI model, CSM-1B, to the public under an Apache 2.0 license. This open-source model, comprising 1 billion parameters, utilizes residual vector quantization (RVQ) to convert text and audio inputs into high-quality speech outputs. By open-sourcing CSM-1B, Sesame aims to foster community engagement and accelerate advancements in AI-driven voice technologies. However, the company advises users against deploying the model for unethical purposes, such as voice mimicry without consent or generating misleading content.
Xbox Introduces ‘Copilot for Gaming’—An AI-Powered Companion for Enhanced Player Experience
Microsoft’s Xbox division has unveiled ‘Copilot for Gaming’, an AI-driven assistant designed to enrich the gaming experience for players. Accessible through the Xbox mobile app, this voice-activated companion offers real-time tips, strategic advice, and personalized recommendations. For instance, in games like Overwatch 2, Copilot can analyze past performances to suggest optimal character selections and strategies. Additionally, it can notify players when friends are online and recommend new titles based on individual preferences. Currently in the testing phase with Xbox Insiders, Microsoft plans to refine Copilot based on user feedback before a broader release.
UiPath Acquires Peak AI to Strengthen Agentic AI Capabilities Amid Economic Uncertainty
UiPath, a leader in robotic process automation, has acquired Peak AI, a UK-based startup specializing in decision-making AI for sectors like retail and manufacturing. This strategic acquisition aims to enhance UiPath’s offerings in agentic AI, providing clients with advanced tools for pricing strategies and inventory management. Despite reporting a 5% increase in fourth-quarter revenue to $424 million, UiPath has adjusted its fiscal year 2026 revenue forecast to $1.525 billion to $1.530 billion, citing global macroeconomic uncertainties. The integration of Peak AI is anticipated to bolster UiPath’s growth trajectory by expanding its AI-driven automation solutions.
Snapchat Introduces AI Video Lenses Powered by In-House Generative Model
Snapchat has unveiled its first-ever AI Video Lenses, leveraging its proprietary generative video model to enhance user experience. These innovative lenses are initially available to subscribers of Snapchat Platinum, the platform’s premium tier priced at $15.99 per month.
The debut includes three distinct lenses: “Raccoon” and “Fox,” which animate virtual animals interacting playfully with users, and “Spring Flowers,” offering a zoom-out effect revealing the user holding a bouquet. Snapchat plans to expand this collection by introducing new AI Video Lenses weekly. This development underscores Snap’s commitment to integrating advanced augmented reality (AR) and artificial intelligence (AI) technologies, aiming to provide creative tools that set it apart from competitors like Instagram and TikTok.
Pentera Secures $60M Funding, Achieving Unicorn Status to Enhance Cybersecurity Training
Pentera, a leader in automated security validation, has announced a $60 million Series D funding round, elevating its valuation to over $1 billion. The Boston-based, Tel Aviv-founded company specializes in simulating network attacks to assess and fortify organizational cybersecurity defenses.
With a remarkable 200% increase in customers, now totaling 1,100 organizations, and a 300% surge in annual recurring revenue over the past four years, Pentera’s growth reflects the escalating demand for robust cybersecurity measures. The latest funding, led by Evolution Equity Partners with participation from Farallon Capital, will be directed towards strategic acquisitions and product development, reinforcing Pentera’s mission to provide comprehensive security solutions in an era of sophisticated cyber threats.
Google DeepMind Unveils Gemini Robotics AI Model to Enhance Robotic Capabilities
Google DeepMind has introduced Gemini Robotics, an advanced AI model that integrates language, vision, and physical action to enable robots to perform complex tasks. Demonstrations showcased robots executing activities such as folding paper and handling objects based on spoken commands, highlighting the model’s potential to revolutionize robotic assistance in various settings. citeturn0news40
The Gemini Robotics-ER variant emphasizes visual and spatial understanding, enhancing embodied reasoning. Designed for adaptability across different hardware platforms, these models aim to empower researchers and developers to advance robotic capabilities. Despite these advancements, considerations regarding the safety and ethical implications of AI-driven robots persist, prompting Google DeepMind to introduce the ASIMOV benchmark to identify and mitigate potentially hazardous behaviors. citeturn0news40
Bria Secures Funding to Develop AI Models Trained on Licensed Data
Bria, a pioneering startup in the AI domain, has successfully secured new funding to advance its development of AI models trained on licensed data. This strategic approach ensures that the AI systems are both ethical and legally compliant, addressing growing concerns about data privacy and intellectual property rights.
By utilizing licensed datasets, Bria aims to set a new standard in the industry, promoting responsible AI development and fostering trust among users and stakeholders. This funding will accelerate Bria’s efforts to innovate in the AI space while adhering to ethical guidelines and legal frameworks.
From groundbreaking AI models and billion-dollar investments to next-gen robotics and AI-powered digital experiences, this week has proven that AI is evolving at an unstoppable pace. The competition is heating up, and the innovations are reshaping industries faster than ever. Whether it’s Google’s AI healthcare updates, Meta’s record-breaking Llama downloads, or OpenAI’s latest integrations, the message is clear—AI is the future, and it’s happening NOW.
Stay tuned for next week’s AI roundup, where we’ll bring you the latest breakthroughs, industry shifts, and must-know updates. Don’t miss out—keep learning, stay informed, and be part of the AI revolution!
